Re: [squid-users] Sizing a pretty Large Squid Install

Mark Pace Balzan wrote:

> At what point do I need to consider clustering ? (Note the above is at one
> single physical location.)

Probably from start. You already know that you need to grow quite a bit
in request rate, and clustering is good for reliability.

> Practically speaking what is the max load seen in the field/production ?
> Does squid break under such heavy load, if running on appropriate
> hardware/memory ?

Squid is a quite big piece of softare, and occationally it crashes (and
restarts shortly thereafter). Also, many people find that hardware quite
often locks up / crashes when running Squid under high load, and also it
has a tendency to stress the VM system of the OS quite badly.. Squid is
a huge process doing lots of random I/O in parallell, causing quite
large disk service times and stress on filesystem buffers and uses many
networking kernel resources, which in when combined stresses the kernel
memory consumption more than usual.

> Is anyone aware of any hardware, kernel o/s or squid issues with
> memory addressing over 1GB physical RAM

Plenty in various OS:es, but you should be run if you run a reasonably
modern OS version.

Squid should not care as long as you have less RAM than the signed
pointer size of your processor..

--
Henrik Nordstrom
Squid Hacker